Following a discussion on irc, we installed the staging db without specifying a collation.
This resulted in configuring en_US.UTF-8 on staging db (or even on the provenance server).
This creates sorting discrepancies down the line [2].
This should be fixed.
The M.O. is to create a db dump, scratch the db, create a new one (with the proper collation), finally restore the dump on new db [3]
[1] https://www.postgresql.org/docs/12/collation.html
[2] P1236
[3] Top of my head following the discussion, this needs proper documentation source link